EMUSIC-L Digest Volume 49, Issue 03 This issue's topics: Anyone using Jam Factory or Markov Processes? jam factory & markov process Markov processes Markov Process (2 messages) Markov processes (3 messages) Your EMUSIC-L Digest moderator is Joe McMahon . You may subscribe to EMUSIC-L by sending mail to listserv@american.edu with the line "SUB EMUSIC-L your name" as the text. The EMUSIC-L archive is a service of SunSite (sunsite.unc.edu) at the University of North Carolina. ------------------------------------------------------------------------ Date: Thu, 18 Feb 1993 15:35:58 GMT From: Clarence Din Subject: Anyone using Jam Factory or Markov Processes? I'm doing a survey paper on Markov Processes in Automated Music Composition and was wondering if any of you actually use Markov Processes in your compositions. I have seen a program called Jam Factory which came out several years ago which use the technique. Has anyone used this program to successfully write "interesting" music? Tell me about your experiences! And, if you have any references on Markov Processes in Automated Music Composition, please direct me to some good sources! I already have Elements of Computer Music by F. R. Moore and some "selected readings"-type book from MIT's Comp Music Journal. Thanks a bunch! --------------------------------------------------------------------------- Clarence K. Din Information Systems Specialist III, Dept of Chemistry, Univ of Pennsylvania din@gradient.cis.upenn.edu --------------------------------------------------------------------------- Too many toys (D-50+M-EX, U-220, SR-16, TurboDuo, SNES, Mac IIsi, HP48SX) and too little time! ------------------------------ Date: Thu, 18 Feb 1993 15:31:21 CST From: Michael Whitten Subject: RE: jam factory & markov process On Thurs. 2/18/93 Clarence Din wrote: >I'm doing a survey paper on Markov Processes in Automated Music >Composition and was wondering if any of you actually use Markov >Processes in your compositions. I have seen a program called >Jam Factory which came out several years ago which use the >technique. Has anyone used this program to successfully write >"interesting" music? Tell me about your experiences! >And, if you have any references on Markov Processes in Automated Music >Composition, please direct me to some good sources! >I already have Elements of Computer Music by F. R. Moore and some >"selected readings"-type book from MIT's Comp Music Journal. >Thanks a bunch! >--------------------------------------------------------------------------- >Clarence K. Din Dammit, Din, I'm a composer - not a programmer! What is a Markov Process? ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ + Michael D. Whitten M-WHIT2@VM1.SPCS.UMN.EDU M-WHIT2@UMINN1 + + "Psst. Hey, Guido. Its all so clear to me now." + ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ------------------------------ Date: Thu, 18 Feb 1993 17:10:18 -0500 From: "Joseph D. McMahon" Subject: Markov processes Michael Whitten writes: > > Dammit, Din, I'm a composer - not a programmer! What is a Markov Process? > Markov processes are interesting little animals. The general idea is to take a set of text, or painting gestures, or musical notes, the bigger the better, and analyze it so as to break it up into "chains" of events. For text, this will normally be words; for music, notes. This "corpus" is analyzed to determine the frequencies at which patterns of events occur. Depending on how correlated the input is to be to the output, you use more or fewer events as context for the next event. The patterns are placed into a table, with the relative frequencies of occurrence associated with them. Random numbers are then used to "walk" down the chains. Adding pseudo-events, such as "beginning of sentence", "end of sentence", and "end of corpus" can be used to give more context to the chains. So if a piece contains patterns of notes where a descending pattern of G, F, and Eb occurs more often than a pattern of G, F, and D, the Markov process will be weighted to randomly selecting the first possibility rather than the latter. Visualize the process thusly: Take a large number of copies of a tape of a musical line. Chop them all up into 4 or 5 note pieces such that you have each note at the beginning of a piece of tape. Now, pick a piece of tape at random. Listen to it. Now find all of the other pieces of tape that start with the same three notes that end the piece that you just listened to. Pick one at random and splice the new two notes onto the piece of tape you had already. Repeat until the resulting tape is "long enough". Certainly this is something you would never physically want to do. Using the computer allows it to build up a table of notes and context and to do this process in real time. --- Joe M. ------------------------------ Date: Fri, 19 Feb 1993 10:38:46 CST From: Michael Whitten Subject: Markov Process On Feb 18, 1993, Joe McMahon responded to my query What is a Markov Process? with: >Markov processes are interesting little animals. The general idea is to > > Joe McMahon Very interesting, Joe. The program 'AlgoRhythms' does exactly that, I believe, and I have used its output on occasion. It is able to utilize user-written 'style-forms' to change the output's character. Your tape analogy reminds me of Turing's single-dimension automata.......are Markov processes related to this? ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ + Michael D. Whitten M-WHIT2@VM1.SPCS.UMN.EDU M-WHIT2@UMINN1 + + "Psst. Hey, Guido. Its all so clear to me now." + ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ------------------------------ Date: Fri, 19 Feb 1993 17:39:29 -0500 From: "Joseph D. McMahon" Subject: Re: Markov Process Michael Whitten writes: > > Very interesting, Joe. The program 'AlgoRhythms' does exactly that, I > believe, and I have used its output on occasion. It is able to utilize > user-written 'style-forms' to change the output's character. Your tape > analogy reminds me of Turing's single-dimension automata.......are Markov > processes related to this? > Could be. We're wandering off into deep waters, and the best I can offer here is a rubber ducky. I'd think it would certainly be possible to program a two-tape Turing machine to do it quite easily; a one-tape version could probably be done too, but would require sufficient storage space to save the context tables. Thank goodness our computational models are (mostly) a lot more sophisticated. :-) --- Joe M. ------------------------------ Date: Fri, 19 Feb 1993 14:31:10 CST From: crispen Subject: Re: Markov processes "Joseph D. McMahon" >Markov processes are interesting little animals. ... >Visualize the process thusly: Take a large number of copies of a tape >of a musical line. Chop them all up into 4 or 5 note pieces such that >you have each note at the beginning of a piece of tape. Now, pick a >piece of tape at random. Listen to it. Now find all of the other >pieces of tape that start with the same three notes that end the >piece that you just listened to. Pick one at random and splice the new >two notes onto the piece of tape you had already. Repeat until the >resulting tape is "long enough". Now it's been many, many years since I've looked at Markov processes (in the context of why they were inadequate as a mechanism to describe speech) but I seem to recollect that they're state machines, in that each link in the Markov chain should be kept ignorant of the information that caused the machine to activate that link. That is, if you're the link that is activated whenever notes A, B and C are generated, that you really don't know anything about what's gone on before. You are programmed to generate one of a series of notes, each with a probability of generation preassigned to it (usually based on the empirical frequency derived from the corpus -- A, B, and C are followed by D x% of the time, by E y% of the time, and so on). The lovely thing about Markov processes is that if you decide beforehand that you'll generate notes based on only one immediately preceding note, you only have to have 12 finite state machines; with 2 preceding notes, you only need 144 fsm's and so on. Each of these little guys flips a coin and decides based on its built-in tables which note to generate next, thereby activating the next link. Or have I got Markov processes confused with something else? +-------------------------------+--------------------------------------+ | Rev. Bob "Bob" Crispen | Most people wouldn't know music if it| | crispen@foxy.boeing.com | came up and bit them on the ass -- FZ| +-------------------------------+--------------------------------------+ ------------------------------ Date: Fri, 19 Feb 1993 17:35:05 -0500 From: "Joseph D. McMahon" Subject: Re: Markov processes No, you've just pointed out an ambiguity in my carefully-crafted analogy. :-) I meant to state specifically that the last three notes are all you care about; on rereading, I imply that (sorta) but don't really come right out and say it. I was too caught up in my impractical, but vivid, physical model. Yes, you are absolutely right, only the ending context makes any difference when choosing the next event. The algorithm doesn't care about how long it's been producing this stuff... On a semi-related note, has anyone ever tried adapting the emacs dissociated-press algorithm to mangling a piece of music? Seems like that could generate some unusual results. --- Joe M. ------------------------------ Date: Sun, 21 Feb 1993 15:34:31 EST From: Derek Burney Subject: Re: Markov processes Hi there, I'm in the process of taking a "performance modelling" course and we just happen to be in the middle of Markov processes. The gentleman who suggested that Markov processes are really finite state machines is quite correct. In fact, they are called "memoryless" because as he correctly stated, whether a state moves from A to B or not, is entirely unrelated to its past behaviour, ie. the length of time spent in A or how it got there in the first place, is irrelevent. Thought I'd chip in my 10c worth... Derek Burney burney@scs.carleton.ca ------------------------------ Date: Fri, 19 Feb 1993 09:15:56 GMT From: Gabriel Landini Subject: Markov processes in composition what about this one: Ait to musical composition employing a random probability system. Olson & Belar. (1961) J Acoust. Soc. Am. 33(9):1163-1170. The problem with the transition probabilities is that allow only short time correlations. You may also want to know about 1/f noise in music. This approach gives you more correlation in the entire piece. Mail me if you want more info. G.Landini@bham.ac.uk ------------------------------ End of the EMUSIC-L Digest ******************************